Optimal Information Retrieval with Complex Utility Functions
Tao Tao, ChengXiang Zhai · Illinois Digital Environment for Access to Learning and Scholarship (University of Illinois at Urbana-Champaign) · 2004
Abstract. Existing retrieval models all attempt to optimize one single utility function, which is often based on the topical relevance of a document with respect to a query. In real applications, retrieval involves more complex utility functions with preferences on several different dimensions. In this paper, we present a general optimization framework for retrieval with complex utility functions. A query language is designed according to this framework to enable users to submit complex queries. We propose an efficient algorithm for retrieval with complex utility functions based on the a-priori algorithm. As a case study, we apply our algorithm to a complex utility retrieval problem in distributed IR. Experiment results show that our algorithm allows for flexible tradeoff between multiple retrieval criteria. Finally, we study the efficiency issue of our algorithm on simulated data. 1